Climate Change Scenarios for the Agricultural and Hydrological Impact Studies Martin Dubrovský (WP 3, 5, 6) Masarykova kolej, Praha ***** 25-27 November 2009 scheme of cc impacts study GCM output obs.wea.series (~15-30 years ) info about - plant genetics - soil properties - growing site - management WG (present climate) B synt.wea.series (present climate) CC scenario WG (future climate) synt.wea.series (future climate) CROP GROWTH MODEL model yields (obs.wea.) C model yields (synt.wea: presence ) D model yields (synt.wea.- future ) A obs. yields A: validation of crop model B: validation of WG C: indirect validation of WG D: climate change impacts (comparison made in terms of means, variability, extremes)
in our climate change impact studies we want to account for the uncertainties - between GCMs (or in RCMs if we would have more of them) - in emission scenarios - in climate sensitivity 11 GCMs Global temperature at SRES-A2 MAGICC (K = clim.sensitivity) IPCC K= 6.0 4.5 3.0 1.5! range of ΔT glob simulated by a set of GCMs is not representative for the! uncertainty in climate sensitivity pattern scaling helps
pattern scaling technique assumption: pattern (spatial and temporal /annual cycle/) is constant, only magnitude changes proportionally to the change in global mean temperature: ΔX(t) = ΔX S x ΔT G (t) where ΔX S = standardised scenario ( = scenario related to ΔT G = 1 C ) a) ΔX S = ΔX [ta-tb] / ΔT G [ta-tb] b) linear regression [x = ΔT G ; y = ΔX] going through zero ΔT G = change in global mean temperature!! ΔT G may be estimated by other means than GCMs!! (e.g. simple climate models /~ MAGICC/) uncertainty in standardised scenario (~between GCM uncertainties)
combining information from 18 / 14 GCMs motivation: to show the multi-model mean/median + uncertainty in a single map step1: results obtained with each of 7 GCMs are re-gridded into 0.5x0.5º grid (~CRU data) step2: median [med(x)] and std [std(x)] from the 18/14 values in each grid box are derived step3 (map): the median is represented by a colour, the shape of the symbol represents value of uncertainty factor Q: Q = std(x) med(x) interpreting the uncertainty: - squares and circles [ std(x) 0.5 * median(x) ] indicate that med(x) differs from 0 at significance level higher than 95% (roughly) - 4-point stars indicate high uncertainty [ std(x) > med(x) ] or: the greater is the proportion of grey (over sea) or black (over land) colour, the lower is the significance, with which the median value differs from 0 multigcm scenarios (standardised) :: TAVG (14 GCMs, SRES-A2) nearly whole Europe: STD(ΔT) < 0.4 * median(δt)
multigcm scenarios (standardised) :: PREC (14 GCMs, SRES-A2)!!! STD > 2*median!!! multigcm scenarios (standardised): TAVG (top ) and PREC (bottom ) (14 GCMs, SRES-A2)
to reflect above uncertainties, we typically use combination of 3 ΔT G x 3 GCMs: uncertainty in ΔT G (modelled by MAGICC): emissions clim.sensitivity high scenario: SRES-A2 4.5 K low scenario: SRES-B1 1.5 K middle scen.: middle 2.5 K uncertainty in pattern: set of GCMs X IPCC-AR3 set + natural variability (day-to-day, year-to-year) is modelled by WG preferred GCMs: HadCM3 NCAR-PCM ECHAM5. + Arpege. now we have outputs from RCM. Q: How to implement it in our impact studies? A. direct output from RCM run at given emission scenario?... NO. It does not reproduce real-climate weather characteristics (annual cycle, temporal structure, ) we do not have enough RCM simulations to account for uncertainties (in emissions, in climate sensitivity, between GCMs) B. using our methodology (WG) but with RCM instead of OBS: present-climate weather series generated by: WG calibrated with RCM output (driven by ERA-40 reanalysis) future climate weather series: WG is modified according to GCM-based scenario CC impact = crop model(future WS) crop model (present WS) results will be presented tomorrow? problem: RCM does not have a satisfactory statistical structure (next slides) solution: debiasing RCM outputs (not available for Cecilian experiments)
RCM vs. OBServations (station data) (in terms of WG parameters) Acknowledgement: European station data for RCM validation were provided by the COST734 project RCM vs OBS: TMAX(July) RCM: Aladin driven by ERA-40 RCM ~ OBS
RCM vs OBS: std(tmax) [JAN] RCM < OBS RCM vs OBS: std(tmax) [JUL] RCM: Aladin driven by ERA-40 RCM < OBS RCM > OBS
RCM vs OBS: PREC(avg.daily sum) [JUL] RCM: Aladin driven by ERA-40 RCM ~ OBS RCM vs OBS: prob(prec>0) [JUL] RCM >> OBS!
RCM vs OBS: cor(tmax,tmin) [JAN] RCM ~ OBS RCM vs OBS: cor(tmax,tmin) [JUL] RCM > OBS (RCM overestimates cor(tx,tn))
RCM vs OBS: autocor(lag=1day) (TMAX) [JAN] RCM < OBS (in most stations) RCM vs OBS: autocor(lag=1day) (TMAX) [JUL] RCM > OBS (RCM overestimates lag-1-cor(tx))
RCM vs OBS: autocor(lag=1day) (TMIN) [JAN] RCM < OBS RCM vs OBS: autocor(lag=1 day) (TMIN) [JUL] RCM > OBS (RCM overestimates lag-1-cor(tn))
conclusion results obtained with available RCM output (using WG calibrated with RCM + GCM-based scenarios): tomorrow (?) the methodology under development: to correct systematic errors in RCM output: WG parameters calibrated from RCM is corrected with (OBS-RCM) bias (RCM = dynamical interpolator WG) to account for the sub-gcm-grid information on scenario from RCM: adding RCM-GCM increment to GCM-based scenarios (interpolated to RCM grid) Acknowledgement: European station data for RCM validation were provided by the COST734 project